Search "AI video production company" and the entire first page is agencies describing themselves. That is not vetting, that is marketing. StudioList sits on the other side of the table: we maintain a directory of AI video studios and we do not produce video, so our only stake is that the studio you pick does good work.
These are the 10 questions we would ask before signing with any AI video studio in July 2026, and what a good answer sounds like for each.
1. Which models are in your stack, and what happens when one of them dies?
This question was theoretical until OpenAI shut down Sora - app dead April 26, 2026, API gone September 24, 2026. Studios that built single-model pipelines are migrating mid-campaign.
Good answer: a named, multi-model stack (for example Seedance for character consistency, Veo for photoreal shots, Kling for volume work) plus a plain statement that the pipeline is model-agnostic. Bad answer: "we use the best AI" with no specifics.
2. Will our footage, products, or brand assets be used to train models?
Enterprise API terms differ wildly from consumer tiers, and open-weight models run entirely on studio hardware. You are allowed to ask exactly where your assets go.
Good answer: a data-handling policy they can show you - which tools see your assets, under which tier, and whether anything is retained after delivery. Bad answer: surprise that anyone asked.
3. Who owns the finished video - legally, not contractually?
The Supreme Court declined to review Thaler v. Perlmutter on March 2, 2026, which settled US law: purely AI-generated footage cannot be copyrighted, and prompting alone is not authorship. Human editing, selection, and arrangement can protect the human-authored components.
Good answer: the studio explains this distinction unprompted, documents human authorship in their workflow, and assigns you everything assignable in the contract. Bad answer: "you own everything" with no mention of the copyright gap.
4. How do you handle human likenesses?
In June 2026, fashion models accused retailer Rainbow Shops of using AI-generated versions of real models without consent or compensation. Likeness disputes are now a live commercial risk, not a hypothetical.
Good answer: written consent and compensation for any real-person likeness, and synthetic-only humans otherwise, with a clearance step in the pipeline. Bad answer: "the AI generates random faces, it is fine."
5. What exactly does a revision round include?
In AI production, "make her turn left instead of right" can mean regenerating a shot dozens of times. Studios price this in, cap it, or lose money - and studios that lose money cut corners.
Good answer: a defined revision policy - what counts as an edit versus a regeneration, how many rounds are included, and what overage costs. Bad answer: "unlimited revisions." Nobody offering unlimited regeneration has done the math, and you do not want to be the client when they do it mid-project.
6. Can you keep our ads compliant with disclosure rules?
New York's synthetic performer disclosure law took effect June 9, 2026: any ad reaching New York audiences that features an AI-generated human likeness needs conspicuous disclosure, at $1,000 per first violation and $5,000 after. The EU AI Act adds labeling duties in Europe. This applies based on where the ad runs, not where anyone is headquartered.
Good answer: they raise disclosure before you do, and can point to campaigns delivered with compliant labeling. Bad answer: "that is your legal team's problem."
7. Which work in your portfolio is actually yours, and can we verify it?
AI showreels are easy to pad - the barrier to generating something impressive-looking is low, and attribution norms are loose.
Good answer: named clients, linkable published work, and ideally third-party validation like festival results - winners of events like Runway AIFF and Project Odyssey are publicly documented. Bad answer: a beautiful reel with zero named, verifiable projects.
8. Who is actually on the team?
The Cannes Lions 2026 takeaway was that craft beat AI novelty, and the Harris Poll presented alongside it found 78% of consumers say AI makes ads feel less authentic and 73% trust an ad less if they suspect it is AI-made. The studios that clear that bar are the ones with directors, editors, and sound designers - people with taste - driving the models.
Good answer: named humans with backgrounds in film, motion, or advertising craft. Bad answer: an anonymous "team of AI artists" with no history you can check.
9. How do you price, and what does the quote license?
Per-project is the norm - most studios on StudioList quote this way. What matters is whether the quote is itemized: deliverable, duration, usage rights, markets, exclusivity, revision rounds.
Good answer: an itemized quote you can compare against our 2026 AI video rate card. Bad answer: one opaque number that turns out to license organic social only, with paid usage sold back to you later.
10. What will AI do badly on this project?
The honesty test. Every working AI video professional knows the current failure modes - long dialogue scenes, precise product physics, brand-exact typography, complex hand interactions. The models improve monthly, but the honest answer is never "nothing."
Good answer: specific limitations plus the workaround plan (hybrid shoots, post-production fixes, scope adjustments). Bad answer: "AI can do anything now." That sentence costs you six weeks and a reshoot.
The red-flag summary
| Red flag | What it predicts |
|---|---|
| Single-model pipeline, no migration plan | Mid-campaign vendor risk (see: Sora) |
| No data-handling answer | Your assets in someone's training set |
| "You own everything," no authorship process | Unprotectable deliverables |
| Unlimited revisions | Corner-cutting when regeneration costs bite |
| No disclosure awareness | Compliance fines land on your ad spend |
| Unverifiable portfolio | You are buying someone else's reel |
Where to start
Every studio on StudioList lists its tools, specializations, verifiable portfolio work, and case studies, so half of these questions are answered before the first call. Browse studios, check festival winners, or get matched free and put the other five questions to studios that already passed the paper test.